Noise and filtering

In [1]:

# to run in google colab
import sys
if 'google.colab' in sys.modules:
    import subprocess
    subprocess.call('apt-get install subversion'.split())
    subprocess.call('svn export https://github.com/YoniChechik/AI_is_Math/trunk/c_02b_filtering_and_resampling/Tour_Eiffel.jpg'.split())

In [2]:

import numpy as np
import cv2
from matplotlib import pyplot as plt

figsize = (10,10)

Get basic image:

In [3]:

def plot_im(img, title):
    plt.figure(figsize=figsize)
    plt.imshow(img)
    plt.title(title)
    plt.xticks([])
    plt.yticks([])
    plt.show()

img = cv2.imread("Tour_Eiffel.jpg")
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
plot_im(img, "original image")

Mean filter

In [4]:

def mean_kernel_smoothing(img, sz):
    kernel = np.ones((sz, sz))/(sz**2)
    dst = cv2.filter2D(img, -1, kernel)

    plot_im(dst, str(sz)+'X'+str(sz)+" mean kernel")


mean_kernel_smoothing(img, 5)
mean_kernel_smoothing(img, 10)
mean_kernel_smoothing(img, 20)

Gaussian filter

In [5]:

def gauss_blur(img, k_sz, sigma=-1, is_plot_kernel=False):
    blur = cv2.GaussianBlur(img, (k_sz, k_sz), sigma)
    plot_im(blur, "gaussian kernel with kernel_size="
            + str(k_sz)+r", $\sigma$=" + str(sigma))
    if is_plot_kernel:
        # sigma=-1 will set the sigma size automatically 
        gauss_ker = cv2.getGaussianKernel(k_sz, sigma)
        plt.figure(figsize=(figsize[0]/2, figsize[1]/2))
        plt.plot(gauss_ker)
        plt.title("corresponding gaussian kernel")
        plt.show()

gauss_blur(img, 5, is_plot_kernel=True)
gauss_blur(img, 21, is_plot_kernel=True)
gauss_blur(img, 21, 1, is_plot_kernel=True)

Median filter

In [6]:

def median_filter(img, k_sz):
    res = cv2.medianBlur(img, k_sz)
    plot_im(res, "median filter with kernel_size="+str(k_sz))


median_filter(img, 5)

Noise addition func

In [7]:

def gauss_noise(image, gauss_var=1000):
    mean = 0
    sigma = gauss_var**0.5
    gauss = np.random.normal(mean, sigma, image.shape)

    res = image + gauss
    noisy = np.clip(res,0,255).astype(np.uint8)
    return noisy

def s_and_p_noise(image, s_p_ratio=0.04):
    # this implementation is not entirely correct because it assumes that
    # only 0 OR 255 values are S&P noise.
    out = np.copy(image)

    # Salt mode
    mask = np.random.rand(image.shape[0],image.shape[1]) <= s_p_ratio/2
    out[mask] = 255

    # Pepper mode
    mask = np.random.rand(image.shape[0],image.shape[1]) <= s_p_ratio/2
    out[mask] = 0
    return out

Gaussian noise tests

In [8]:

np.random.seed(1234)

gauss_noise_im = gauss_noise(img, gauss_var=70)
plt.figure(figsize=figsize)
plt.imshow(gauss_noise_im)
plt.title('original image + gaussian noise')
plt.show()

In [9]:

gauss_blur(gauss_noise_im, 3)
gauss_blur(gauss_noise_im, 5)
gauss_blur(gauss_noise_im, 11)
median_filter(gauss_noise_im, 5)

Salt and pepper noise test

In [10]:

s_p_noise_im = s_and_p_noise(img, s_p_ratio=0.04)
plt.figure(figsize=figsize)
plt.imshow(s_p_noise_im)
plt.title('original image + s&p noise')
plt.show()

In [11]:

gauss_blur(s_p_noise_im, 5)
median_filter(s_p_noise_im, 3)
median_filter(s_p_noise_im, 5)

In [ ]: